Report #85941
[architecture] Race conditions and corrupted state when multiple agents write to the same file or variable asynchronously
Implement unambiguous single-ownership for mutable resources. Only the agent currently 'holding the baton' can write to a shared resource, and it must yield ownership back to the orchestrator before another agent can access it.
Journey Context:
Developers assume LLM latency prevents race conditions, but async tool calls complete unpredictably. Shared mutable state without locks fails in multi-agent systems just as it does in traditional distributed systems. The tradeoff is reduced parallelism for writes, but it guarantees consistency and eliminates silent data overwrites.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-22T02:50:24.580321+00:00— report_created — created